You’ve probably seen a fair bit of Fermi coverage around the interwebs today, and might even be wondering why we’ve not posted something. We were at the same briefing as these other sites, after all. If you’ve read the coverage elsewhere carefully, you’ll notice that there’s the odd dig at not seeing benchmarks, or hardware, from Nvidia and that’s why we’ve not posted a big extravaganza as well as Keir Graham’s mod
While the briefing last Sunday was very informative about Nvidia's thoughts on a variety of topics (not just Fermi), there was little opportunity to test Fermi hardware. Nvidia is just getting the first chips back from the fab, and is still assessing the yields and quality of the silicon in order to set the final clocks, so we still think the actual launch is six to eight weeks away. That means that any article we posted today would essentially say, ‘Nvidia thinks forthcoming Nvidia product will be great’ with the inference, ‘so hold off on buying a HD 5850 or 5870.’
That just doesn’t seem fair – sure, if you can wait a couple of months then do so, but if you need an upgrade now, those cards are very good (if you can find stock). After all, there was no mention in the briefing of how much a Fermi card will cost.
But I don’t want to bash Nvidia too much in this blog, instead I wanted to talk about the atmosphere of those briefings last week. Despite a few self-critical and honest comments regarding Fermi’s lateness and how much Nvidia would have liked to have had a product when Windows 7 launched, the general mood was one of calm enthusiasm and quiet pride. It was the feeling you get handing in a piece of work that you’ve absolutely nailed – that essay that you just know you aced, the report you’ve been working on for weeks that’s going to get you noticed by Them Upstairs, the mod you drop onto the forums that you know will get even the pros in a fluster. Nvidia, I think, reckons it’s on to an absolute winner with Fermi.
Whether this is true remains to be seen – even if the claims of being twice as fast as a GTX 285 and having eight times the geometry processing power are true, the price of the 3bn transistor GPU could still make a Fermi uncompetitive. With clock speeds yet to be defined, and only a couple of Nvidia-approved tests to go on, we’ll also avoid making any assumptions and postulations of how much faster or slower a GF100-equipped card will be than Radeon HD 5870 for the moment.
While Nvidia had a lot of positive things to say about Fermi, we’ll wait until we have hardware to test in order to temper that enthusiasm with our own findings and objective analysis. We’re therefore using the Fermi briefing of last Sunday to get more of a head start on our technical understanding than we’ve been used to in the past.